Early vision segments the retinal image into objects, represents these objects in a manner in which they can be recognized, and guides attention to portions of the visual scene that require further scrutiny. Individual features such as lines and edges play a role in these processes, but often the statistics of visual images are at least as important. For example, in complex images, including naturalistic ones, only a small number of contrast contours represent object boundaries, and consequently objects are more reliably defined by discontinuities in image statistics, rather than by isolated features. Despite the impressive ability of the visual system to analyze scene statistics, previous work (including our studies in current and previous funding periods) indicates that this statistical processing is limited and specific. Motivated by these considerations, we propose four sets of experiments to define the mechanisms and limits of statistical processing of visual images, and how these processes interact with image segmentation, representation in visual working memory, and guidance of attention. Based on theoretical and functional considerations, we consider three classes of visual statistics: statistics that are available at the cortical input (such as luminance and contrast), statistics that require cortical analysis but only within local regions (higher-order elements of form), and statistics that require more long-range analysis (such as bilateral symmetry). The proposed experimental design will allow us to compare these modalities on a common footing, and the proposed modeling approach will determine how processing is shaped by detection of statistical elements, and by pooling of statistical information. The planned research will also lead to insights into how top-down influences such as attention modulate the computations carried out in early visual areas. There are two broad hypotheses: A. The kinds of statistics used by early vision are determined by the spatial selectivity of visual processing, rather than considerations related to formal information content. B. The capacity for processing of visual statistics, though ultimately constrained by computational resources, is more typically tuned to, or modulated by, task demands.